bbreorder odity

[Dan Nicolaescu <dann at godzilla dot ICS dot UCI dot EDU>]

Hi!

Compiling the following 
(stolen from  http://gcc.gnu.org/projects/optimize.html):

int or (int a, int b) { return (a || b); }

with the mainline GCC with -O2 on linux-x86 I get:

or:
	pushl	%ebp
	movl	%esp, %ebp
	movl	8(%ebp), %ecx
	xorl	%eax, %eax
	testl	%ecx, %ecx
	je	.L5
.L4:
	movl	$1, %eax
.L3:
	popl	%ebp
	ret
	.p2align 4,,7
.L5:
	movl	12(%ebp), %edx
	testl	%edx, %edx
	jne	.L4
	jmp	.L3
       ^^^^^

Should it reorder block here? It adds a jump to one of the paths, and
that is the jump to the epilogue. Shouldn't the epilogue be duplicated
here instead of using a jump? 


with -O2 -fno-reorder-blocks

or:
	pushl	%ebp
	movl	%esp, %ebp
	movl	8(%ebp), %ecx
	xorl	%eax, %eax
	testl	%ecx, %ecx
	jne	.L4
	movl	12(%ebp), %edx
	testl	%edx, %edx
	je	.L3
.L4:
	movl	$1, %eax
.L3:
	popl	%ebp
	ret


The odd thing is that the epilogue is duplicated when using 
-O2 -fomit-frame-pointer

or:
	movl	4(%esp), %ecx
	xorl	%eax, %eax
	testl	%ecx, %ecx
	je	.L5
.L4:
	movl	$1, %eax
	ret
	.p2align 4,,7
.L5:
	movl	8(%esp), %edx
	testl	%edx, %edx
	jne	.L4
	ret

I don't know if any of these are real problems, but I thought they are
worth mentioning.